Infrared small target detection based on density peaks searching and weighted multi-feature local difference

作     者：JI Bin FAN Pengxiang WANG Mengli LIU Yang XU Jiafeng 

作者机构：School of Computer Science and Technology Anhui University of Technology 

出 版 物：《Optoelectronics Letters》 (光电子快报(英文))

年 卷 期：2024年

学科分类：08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程] 

基　　金：supported by the National Natural Science Foundation of China (No. 52205548) 

摘      要：In order to solve the problems of unknown target size， fuzzy edges， background interference and low contrast in infrared small target detection， this paper proposes an infrared small target detection method based on density peaks searching and weighted multi-feature local difference. Firstly， for the interference problem of high brightness clutter on the density peaks searching to obtain candidate targets， this paper adopts the improved high-boost filter for preprocessing， which effectively eliminates the background clutter and high brightness background interference in the infrared image， and improves the probability of the true target being captured in the density peaks searching process. Secondly， triple-layer windows are used to extract features from the area around the candidate target， which solves the problem of uncertain size of small targets. Additionally， by calculating the multiple feature local differences between triple-layer windows， the problem of blurred edges and low contrast for the targets are also resolved. In order to balance the contribution degree of different feature local differences to target detection， this paper introduces the intra-class distance to calculate the weight of each feature local differences， realizes the weighted fusion of multi-feature local differences， obtains the weighted multi-feature local difference of each candidate target， and extracts the real target from the candidate target through the interquartile range. Experiments on many different types of infrared image sequences consisting of publicly available datasets such as SIRST and IRSTD-1K show that the method proposed in this paper is applicable to a wide range of complex types with good robustness and detection performance.